OPERATOR ALGEBRAS FOR ANALYTIC VARIETIES

OPERATOR ALGEBRAS FOR ANALYTIC VARIETIES KENNETH R. DAVIDSON, CHRISTOPHER RAMSEY, AND ORR MOSHE SHALIT Dedicated to the memory of William B. Arveson Abstract. We study the isomorphism problem for the multiplier algebras of irreducible complete Pick kernels. These are precisely the restrictions M V of the multiplier algebra M of Drury-Arveson space to a holomorphic subvariety V of the unit ball. We find that M V is completely isometrically isomorphic to M W if and only if W is the image of V under a biholomorphic automorphism of the ball. In this case, the isomorphism is unitarily implemented. This is then strengthend to show that every isometric isomorphism is completely isometric. The problem of characterizing when two such algebras are (algebraically) isomorphic is also studied. When V and W are each a finite union of irreducible varieties and a discrete variety, an isomorphism between M V and M W determines a biholomorphism (with multiplier coordinates) between the varieties; and the isomorphism is composition with this function. These maps are automatically weak- continuous. We present a number of examples showing that the converse fails in several ways. We discuss several special cases in which the converse does hold particularly, smooth curves and Blaschke sequences. We also discuss the norm closed algebras associated to a variety, and point out some of the differences. 1. Introduction In this paper, we study operator algebras of multipliers on reproducing kernel Hilbert spaces associated to analytic varieties in the unit ball of C d. The model is the multiplier algebra M d of the Drury-Arveson space, a.k.a. symmetric Fock space. The generators, multiplication by coordinate functions, form a d-tuple which is universal for commuting row contractions [8]. The Hilbert space is a reproducing kernel Hilbert space which is a complete Nevanlinna-Pick kernel [15]; and in fact when 2010 Mathematics Subject Classification. 47L30, 47A13, 46E22. Key words and phrases. Non-selfadjoint operator algebras, reproducing kernel Hilbert spaces. 1

2 K.R. DAVIDSON, C. RAMSEY, AND O.M. SHALIT d = is the universal complete NP kernel [1]. For these reasons, this space and its multiplier algebra have received a lot of attention in recent years. In this paper, we are concerned with multipliers on subspaces of Drury-Arveson space spanned by the kernel functions they contain. By results in [14], these operator algebras are also complete quotients of M d by wot-closed ideals. The zero set is always an analytic variety V in the ball, and the multiplier algebra M V is a space of holomorphic functions on V. The main question that we address is when two such algebras are isomorphic. We find first that two such algebras M V and M W are completely isometrically isomorphic if and only if there is a biholomorphic automorphism of the ball that carries V onto W. In this case, the isomorphism is unitarily implemented. The question of algebraic isomorphism (which implies continuous algebraic isomorphism because the algebras are semisimple) is much more subtle. In an earlier paper [16], the authors considered the case of homogeneous varieties. We showed, under some extra assumptions on the varieties, that the algebras are isomorphic if and only if there is a biholomorphic map of one variety onto the other. In a recent paper, Michael Hartz [20] was able to establish this result in complete generality. In this paper, we establish a special case of what should be the easy direction, showing that an isomorphism determines a biholomorphism of V onto W. This turns out to be rather subtle, and we need to restrict our attention to the case in which the varieties are a finite union of irreducible varieties and a discrete variety. The isomorphism is just composition with this biholomorphism. These methods also allow us to show that an isometric isomorphism is just composition with a conformal automorphism of the ball, and thus is completely isometric and unitarily implemented. Some counterexamples show that a biholomorphism between varieties does not always yield an isomorphism of the multiplier algebras. We discuss a number of cases where we can establish the desired converse. Arias and Latrémolière [5] have an interesting paper in which they study certain operator algebras of this type in the case where the variety is a countable discrete subset of the unit disc which is the orbit of a point under the action of a Fuchsian group. They establish results akin to ours in the completely isometric case using rather different methods.

OPERATOR ALGEBRAS FOR ANALYTIC VARIETIES 3 2. Reproducing kernel Hilbert spaces associated to analytic varieties Basic notation. Let Hd 2 be Drury-Arveson space (see [8]). H2 d is the reproducing kernel Hilbert space on B d, the unit ball of C d, with kernel functions 1 k λ (z) = for z, λ B d. 1 z, λ We also consider the case d =, and then C d is understood as l 2. We denote by M d the multiplier algebra Mult(H 2 d ) of H2 d. Let Z 1,..., Z d denote multiplication by the coordinate functions on H 2 d, given by (Z i h)(z) = z i h(z) for i = 1,..., d. Let A d denote the norm closed algebra generated by I, Z 1,..., Z d. By [8, Theorem 6.2], A d is the universal (norm-closed) unital operator algebra generated by a commuting row contraction (see also [24]). We write F = F(E) for the full Fock space F = C E (E E) (E E E)..., where E is a d-dimensional Hilbert space. Fix an orthonormal basis {e 1,..., e d } for E. On F, we have the natural shift operators L 1,..., L d given by L j e i1 e ik = e j e i1 e ik for 1 j d. The non-commutative analytic Toeplitz algebra L d is defined to be the wot-closed algebra generated by L 1,..., L d. When d is understood, we may write A, M, L and H 2 instead of A d, M d, L d and H 2 d. The RKHS of a variety. For our purposes, an analytic variety will be understood as the common zero set of a family of H 2 functions. If F is a subset of H 2 d, considered as functions on B d, let V (F ) := {λ B d : f(λ) = 0 for all f F }. Propositions 2.1 and 2.2 below both show that there is not much loss of generality in taking this as our definition in this context. In particular, if f M d, then M f 1 = f is a function in Hd 2. So the zero set of a set of multipliers is an analytic variety. Define J V = {f M : f(λ) = 0 for all λ V }. Observe that J V is a wot-closed ideal in M.

4 K.R. DAVIDSON, C. RAMSEY, AND O.M. SHALIT Proposition 2.1. Let F be a subset of H 2, and let V = V (F ). Then V = V (J V ) = {λ B d : f(λ) = 0 for all f J V }. Proof. Obviously V V (J V ). For the other inclusion, recall that [2, Theorem 9.27] states that a zero set of an H 2 function is a weak zero set for M (i.e. the intersection of zero sets of functions in M). Since V is the intersection of zero sets for H 2, it is a weak zero set for M; i.e., there exists a set S M such that V = V (S). Now, S J V, so V = V (S) V (J V ). Given the analytic variety V, we define a subspace of H 2 d by F V = span{k λ : λ V }. The Hilbert space F V is naturally a reproducing kernel Hilbert space of functions on the variety V. One could also consider spaces of the form F S = span{k λ : λ S} where S is an arbitrary subset of the ball. The following proposition shows that there is no loss of generality in considering only analytic varieties generated by H 2 functions. Proposition 2.2. Let S B d. Let J S denote the set of multipliers vanishing on S, and let I S denote the set of all H 2 functions that vanish on S. Then F S = F V (IS ) = F V (JS ). Proof. Clearly F S F V (IS ). Let f FS. Then f(x) = 0 for all x S; so f I S. Hence by definition, f(z) = 0 for all z V = V (I S ); whence f FV (I S ). Therefore F S = F V (IS ). The extension to zero sets of multipliers follows again from [2, Theorem 9.27]. Remarks 2.3. In general, it is not true that V (I S ) is equal to the smallest analytic variety in the classical sense containing S B d. In fact, by Weierstrass s Factorization Theorem, every discrete set Z = {z n } n=1 in D is the zero set of some holomorphic function on D. However, if the sequence Z is not a Blaschke sequence, then there is no nonzero function in H 2 that vanishes on all of it. So here I Z = {0}, and therefore V (I Z ) = D. One very nice property of classical varieties is that the definition is local. Because our functions must be multipliers, a strictly local definition does not seem to be possible. However one could consider the following variant: V is a variety of for each point λ B d, there is an ε > 0 and a finite set f 1,..., f n in M d so that b ε (λ) V = {z b ε (λ) : 0 = f 1 (z) = = f n (z)}. We do not know if every variety of this type is actually the intersection of zero sets.

OPERATOR ALGEBRAS FOR ANALYTIC VARIETIES 5 In particular, we will say that a variety V is irreducible if for any regular point λ V, the intersection of zero sets of all multipliers vanishing on a small neighbourhood V b ε (λ) is exactly V. However we do not know whether an irreducible variety is connected. A local definition of our varieties would presumably clear up this issue. Ideals and invariant subspaces. We will apply some results of Davidson-Pitts [14, Theorem 2.1] and [15, Corollary 2.3] to the commutative context. In the first paper, a bijective correspondence is established between the collection of wot-closed ideals J of L d and the complete lattice of subspaces which are invariant for both L d and its commutant R d, the algebra of right multipliers. The pairing is just the map taking an ideal J to its closed range µ(j) := JF. The inverse map takes a subspace N to the ideal J of elements with range contained in N. In [15, Theorem 2.1], it is shown that the quotient algebra L d /J is completely isometrically isomorphic and wot-homeomorphic to the compression of L d to µ(j). In particular, [15, Corollary 2.3] shows that the multiplier algebra M d is completely isometrically isomorphic to L d /C, where C is the wot-closure of the commutator ideal of L d. In particular, µ(c) = Hd 2. It is easy to see that there is a bijective correspondence between the lattice of wot-closed ideals Id(M d ) of M d and the wot-closed ideals of L d which contain C. Similarly there is a bijective correspondence between invariant subspaces N of M d and invariant subspaces of L d which contain µ(c) = Hd 2. Since the algebra M d is abelian, it is also the quotient of R d by its commutator ideal, which also has range Hd 2. So the subspace N Hd 2 is invariant for both L d and R d. Therefore an application of [14, Theorem 2.1] yields the following consequence: Theorem 2.4. Define the map α : Id(M d ) Lat(M d ) by α(j) = J1. Then α is a complete lattice isomorphism whose inverse β is given by β(n) = {f M d : f 1 N}. Moreover [15, Theorem 2.1] then yields: Theorem 2.5. If J is a wot-closed ideal of M d with range N, then M d /J is completely isometrically isomorphic and wot-homeomorphic to the compression of M d to N. The multiplier algebra of a variety. The reproducing kernel Hilbert space F V comes with its multiplier algebra M V = Mult(F V ). This is the algebra of all functions f on V such that fh F V for all h F V. A standard argument shows that each multiplier determines a bounded

6 K.R. DAVIDSON, C. RAMSEY, AND O.M. SHALIT linear operator M f B(F V ) given by M f h = fh. We will usually identify the function f with its multiplication operator M f. We will also identify the subalgebra of B(F V ) consisting of the M f s and the algebra of functions M V (endowed with the same norm). One reason to distinguish f and M f is that sometimes we need to consider the adjoints of the operators M f. The distinguishing property of these adjoints is that Mf k λ = f(λ)k λ for λ V, in the sense that if A k λ = f(λ)k λ for λ V, then f is a multiplier. The space F V is therefore invariant for the adjoints of multipliers; and hence it is the complement of an invariant subspace of M. Thus an application of Theorem 2.5 and the complete Nevanlinna-Pick property yields: Proposition 2.6. Let V be an analytic variety in B d. Then M V = {f V : f M}. Moreover the mapping ϕ : M M V given by ϕ(f) = f V induces a completely isometric isomorphism and wot-homeomorphism of M/J V onto M V. For any g M V and any f M such that f V = g, we have M g = P FV M f FV. Given any F M k (M V ), one can choose F M k (M) so that F V = F and F = F. Proof. Theorem 2.5 provides the isomorphism between M/J V and the restriction of the multipliers to N where N = J V 1. Since J V vanishes on V, if f J V, we have M f h, k λ = h, M f k λ = 0 for all λ V and h H 2 d. So N is orthogonal to F V. Conversely, if M f has range orthogonal to F V, the same calculation shows that f J V. Since the pairing between subspaces and ideals is bijective, we deduce that N = FV. The mapping of M/J V into M V is given by compression to F V by sending f to P FV M f FV. It is now evident that the restriction of a multiplier f in M to V yields a multiplier on F V, and that the norm is just f + J V = P FV M f FV. We need to show that this map is surjective and completely isometric. This follows from the complete Nevanlinna-Pick property as in [15, Corollary 2.3]. Indeed, if F M k (M V ) with F = 1, then standard computations show that if λ 1,..., λ n lie in V, then [ (Ik F (λ j )F (λ i ) ) ] k, k λi λj is positive semidefinite. By [15], this implies that there is a matrix multiplier F M k (M) with F = 1 such that F V = F. n n

OPERATOR ALGEBRAS FOR ANALYTIC VARIETIES 7 We can argue as in the previous subsection that there is a bijective correspondence between wot-closed ideals of M V and its invariant subspaces: Corollary 2.7. Define the map α : Id(M V ) Lat(M V ) by α(j) = J1. Then α is a complete lattice isomorphism whose inverse β is given by β(n) = {f M V : f 1 N}. Remark 2.8. By Theorem 4.2 in [1], every irreducible complete Nevanlinna-Pick kernel is equivalent to the restriction of the kernel of Drury- Arveson space to a subset of the ball. It follows from this and from the above discussion that every multiplier algebra of an irreducible complete Nevanlinna-Pick kernel is completely isometrically isomorphic to one of the algebras M V that we are considering here. Remark 2.9. By the universality of Z 1,..., Z d [8], for every unital operator algebra B that is generated by a pure commuting row contraction T = (T 1,..., T d ), there exists a surjective unital homomorphism ϕ T : M B that gives rise to a natural functional calculus f(t 1,..., T d ) = ϕ T (f) for f M. So it makes sense to say that a commuting row contraction T annihilates J V if ϕ T vanishes on J V. By Proposition 2.6, we may identify M V with the quotient M/J V, thus we may identify M V as the universal wot-closed unital operator algebra generated by a pure commuting row contraction T = (T 1,..., T d ) that annihilates J V. 3. The character space of M V If A is a Banach algebra, denote the set of multiplicative linear functionals on A by M(A); and endow this space with the weak- topology. We refer to elements of M(A) as characters. Note that all characters are automatically unital and continuous with norm one. When A is an operator algebra, characters are. When V is an analytic variety in B d, we will abuse notation and let Z 1,..., Z d also denote the images of the coordinate functions Z 1,..., Z d of M in M V. Since [ Z 1,..., Z d ] is a row contraction, ( ρ(z 1 ),..., ρ(z d ) ) 1 for all ρ M(M V ). The map π : M(M V ) B d given by π(ρ) = (ρ(z 1 ),..., ρ(z d ))

8 K.R. DAVIDSON, C. RAMSEY, AND O.M. SHALIT is continuous as a map from M(M V ), with the weak- topology, into B d (endowed with the weak topology in the case d = ). We define V M = π((m(m V )). Since π is continuous, V M is a (weakly) compact subset of B d. For every λ V M, the fiber over λ is defined to be the set π 1 (λ) in M(M V ). We will see below that V V M, and that over every λ V the fiber is a singleton. Every unital homomorphism ϕ : A B between Banach algebras induces a mapping ϕ : M(B) M(A) by ϕ ρ = ρ ϕ. If ϕ is a continuous isomorphism, then ϕ is a homeomorphism. We will see below that in many cases a homomorphism ϕ : M V M W gives rise to an induced map ϕ : M(M W ) M(M V ) which has additional structure. The most important aspect is that ϕ restricts to a holomorphic map from W into V. The weak- continuous characters of M V. In the case of M d, the weak- continuous characters coincide with the point evaluations at points in the open ball [6, 13] ρ λ (f) = f(λ) = fν λ, ν λ for λ B d, where ν λ = k λ / k λ. The fibers over points in the boundary sphere are at least as complicated as the fibers in M(H ) [14], which are known to be extremely large [22]. As a quotient of a dual algebra by a weak- closed ideal, the algebra M V inherits a weak- topology. As an operator algebra concretely represented on a reproducing kernel Hilbert space, M V also has the weak-operator topology (wot). In [16, Lemma 11.9] we observed that, as is the case for the free semigroup algebras L d [13], the weak-operator and weak- topologies on M V coincide. The setting there was slightly different, but the proof remains the same. It relies on the observation [7] that M V has property A 1 (1). coin- Lemma 3.1. The weak- and weak-operator topologies on M V cide. Proposition 3.2. The wot-continuous characters of M V can be identified with V. Moreover, V M B d = V. The restriction of each f M V to V is a bounded holomorphic function. Proof. As M V is the multiplier algebra of a reproducing kernel Hilbert space on V, it is clear that for each λ V, the evaluation functional ρ λ (f) = f(λ) = fν λ, ν λ

OPERATOR ALGEBRAS FOR ANALYTIC VARIETIES 9 is a wot-continuous character. On the other hand, the quotient map from the free semigroup algebra L onto M V is weak-operator continuous. Thus, if ρ is a wotcontinuous character of M V, then it induces a wot-continuous character on L by composition. Therefore, using [14, Theorem 2.3], we find that ρ must be equal to the evaluation functional ρ λ at some point λ B d. Moreover ρ λ annihilates J V. By Proposition 2.1, the point λ lies in V. If ρ is a character on M V such that π(ρ) = λ B d, then again it induces a character ρ on L with the property that ρ(l 1,..., L d ) = λ. By [14, Theorem 3.3], it follows that ρ is wot-continuous and coincides with point evaluation. Hence by the previous paragraph, λ belongs to V. So V M B d = V. Therefore π : π 1 (V ) V is seen to be a homeomorphism between π 1 (V ) endowed with the weak- topology and V with the (weak) topology induced from B d. By Proposition 2.6, M V is a quotient of M, and the map is given by restriction to V. Hence f is a bounded holomorphic function by [14, Theorem 3.3] or [8, Proposition 2.2]. Thus the character space M(M V ) consists of V and M(M V ) \ V, which we call the corona. By definition, the corona is fibered over V M \ V, and by the above proposition this latter set is contained in B d. 4. Completely isometric isomorphisms The automorphisms of M arise as composition with an automorphism of the ball (i.e., a biholomorphism of the ball onto itself). This can be deduced from [14, Section 4], or alternatively from Theorems 3.5 and 3.10 in [26]. In [16, Section 9], we wrote down the explicit form of the unitaries on H 2 that implement these automorphisms. We will use these unitaries to construct unitarily implemented, completely isometric isomorphisms of the multiplier algebras that we are studying. In addition, we will show that all completely isometric isomorphisms of these algebras arise in this way. Proposition 4.1. Let V and W be varieties in B d. Let F be an automorphism of B d that maps W onto V. Then f f F is a unitarily implemented completely isometric isomorphism of M V onto M W ; i.e. M f F = UM f U. The unitary U is the linear extension of the map U k w = c w k F (w) for w W,

10 K.R. DAVIDSON, C. RAMSEY, AND O.M. SHALIT where c w = (1 F 1 (0) 2 ) 1/2 k F 1 (0)(w). Proof. Let F be such an automorphism, and set α = F 1 (0). By [16, Theorem 9.2], the unitary map U B(H 2 ) is given by Uh = (1 α 2 ) 1/2 k α (h F ) for h H 2. As F (W ) = V, U takes the functions in H 2 that vanish on V to the functions in H 2 that vanish on W. Therefore it takes F V onto F W. Let us compute U. For h H 2 and w W, we have h, U k w = Uh, k w = (1 α 2 ) 1/2 k α (h F ), k w = (1 α 2 ) 1/2 k α (w) h(f (w)) = h, c w k F (w), where c w = (1 F 1 (0) 2 ) 1/2 k F 1 (0)(w). Thus U k w = c w k F (w). Note that since U is a unitary, c w = k w / k F (w). Finally, we show that conjugation by U implements the isomorphism between M V and M W given by composition with F. Observe that Uc w k F (w) = k w. For f M V and w W, UM f U k w = UM f c w k F (w) = f(f (w))uc w k F (w)) = (f F )(w)k w. Therefore f F is a multiplier on F W and M f F = UM f U. Now we turn to the converse. Lemma 4.2. Let V B d and W B d be varieties. Let ϕ be a unital, completely contractive algebra isomorphism of M V into M W. Then there exists a holomorphic map F : B d B d such that (1) F (W ) V. (2) F W = ϕ W. (3) the components f 1,..., f d of F form a row contraction of operators in M d. (4) ϕ is given by composition with F, that is ϕ(f) = f F for f M V. Proof. Consider the image of the coordinate functions Z i in M V. As ϕ is completely contractive, Proposition 2.6 shows that [ ϕ(z 1 )... ϕ(z d ) ] is the restriction to W of a row contractive multiplier F = [ ] f 1,..., f d with coefficients in M d. As F is contractive as a multiplier, it is also contractive in the sup norm. Moreover, since ϕ is injective, the f i and F are non-constant holomorphic functions. Therefore F must have range in the open ball B d.

OPERATOR ALGEBRAS FOR ANALYTIC VARIETIES 11 Fix λ W, and let ρ λ be the evaluation functional at λ on M W. Then ϕ (ρ λ ) is a character in M(M V ). We want to show that it is also an evaluation functional. Compute [ϕ (ρ λ )](Z i ) = Z i (ϕ (ρ λ )) = ρ λ (ϕ(z i )) = ϕ(z i )(λ). So ϕ (ρ λ ) lies in the fiber over (ϕ(z 1 )(λ),..., ϕ(z d )(λ)) = F (λ). This is in the interior of the ball. By Proposition 3.2, ϕ (ρ λ ) is the point evaluation functional ρ F (λ) and F (λ) V. We abuse notation by saying that ϕ (ρ λ ) V. Finally, for every f M V and every λ W, Therefore ϕ(f) = f F. ϕ(f)(λ) = ρ λ (ϕ(f)) = ϕ (ρ λ )(f) = ρ F (λ) (f) = (f F )(λ). Lemma 4.3. Let 0 V B d and 0 W B d be varieties. Let ϕ : M V M W be a completely isometric isomorphism such that ϕ ρ 0 = ρ 0. Then there exists an isometric linear map F of B d span W onto B d span V such that F (W ) = V, F (0) = 0 and F W = ϕ. Proof. By making d smaller, we may assume that C d = span V. Similarly, we may assume C d = span W. By Lemma 4.2 applied to ϕ, there is a holomorphic map F of B d into B d that implements ϕ. Thus F (W ) V and F (0) = 0. By the same lemma applied to ϕ 1, there is a holomorphic map G of B d into B d that implements (ϕ 1 ). Hence G(V ) W and G(0) = 0. Now, ϕ and (ϕ 1 ) are inverses of each other. Therefore F G V and G F W are the identity maps. Let H = F G. Then H is a holomorphic map of B d into itself, such that H V is the identity. In particular H(0) = 0. By [27, Theorem 8.2.2], the fixed point set of H is an affine set equal to the fixed point set of H (0) in B d. Therefore H is the identity on B d since C d = span V. Applying the same reasoning to G F, we see that F is a biholomorphism of B d onto B d such that F (W ) = V. In particular, d = d. It now follows from a theorem of Cartan [27, Theorem 2.1.3] that F is a unitary linear map. Now we combine these lemmas to obtain the main result of this section. If V B d and W B d are varieties, then we can consider them both as varieties in B max(d,d ). Therefore, we may assume that d = d. This does not change the operator algebras. See [16, Remark 8.1].

12 K.R. DAVIDSON, C. RAMSEY, AND O.M. SHALIT Theorem 4.4. Let V and W be varieties in B d. Then M V is completely isometrically isomorphic to M W if and only if there exists an automorphism F of B d such that F (W ) = V. In fact, every completely isometric isomorphism ϕ : M V M W arises as composition ϕ(f) = f F where F is such an automorphism. In this case, ϕ is unitarily implemented by the unitary sending the kernel function k w F W to a scalar multiple of the kernel function k F (w) F V. Proof. If there is such an automorphism, then the two algebras are completely isometrically isomorphic by Proposition 4.1; and the unitary is given explicitly there. Conversely, assume that ϕ is a completely isometric isomorphism of M V onto M W. By Lemma 4.2, ϕ maps W into V. Pick a point w 0 W and set v 0 = ϕ (w 0 ). By applying automorphisms of B d that move v 0 and w 0 to 0 respectively, and applying Proposition 4.1, we may assume that 0 V and 0 W and ϕ (0) = 0. Now we apply Lemma 4.3 to obtain an isometric linear map F of the ball B d span W onto the ball B d span V such that F W = ϕ. In particular, span W and span V have the same dimension. (Caveat: this is only true in the case that both V and W contain 0.) We may extend the definition of F to a unitary map on C d, and so it extends to a biholomorphism of B d. Now Proposition 4.1 yields a unitary which implements composition by ϕ. By Lemma 4.2, every completely isometric isomorphism ϕ is given as a composition by ϕ. So all maps have the form described. There is a converse to Lemma 4.2, which may provide an alternative proof for one half of Theorem 4.4. Arguments like the following are not uncommon in the theory of RKHS; see for example [23, Theorem 5]. Proposition 4.5. Let V B d and W B d be varieties. Suppose that there exists a holomorphic map F : B d B d that satisfies F (W ) V, such that the components f 1,..., f d of F form a row contraction of operators in M d. Then the map given by composition with F ϕ(f) = f F for f M V yields a unital, completely contractive algebra homomorphism of M V into M W. Proof. Composition obviously gives rise to a unital homomorphism, so all we have to demonstrate is that ϕ is completely contractive. We make use of the complete NP property of these kernels.

OPERATOR ALGEBRAS FOR ANALYTIC VARIETIES 13 Let G M k (M V ) with G 1. Then for any N points w 1,..., w N in W, we get N points F (w 1 ),..., F (w N ) in V. The fact that G 1 implies that the N N matrix with k k matrix entries [ ] Ik (G F )(w i )(G F )(w j ) 0. 1 F (w i ), F (w j ) N N Also, since F 1 as a multiplier on F W, we have that [ ] 1 F (wi ), F (w j ) 0. 1 w i, w j N N Therefore the Schur product of these two positive matrices is positive: [ ] Ik (G F )(w i )(G F )(w j ) 0. 1 w i, w j N N Now the complete NP property yields that G F is a contractive multiplier in M k (M W ). 5. Isomorphisms of algebras and biholomorphisms We turn now to the question: when does there exist an (algebraic) isomorphism between M V and M W? This problem is more subtle, and we frequently need to assume that the variety sits inside a finite dimensional ambient space. Even the construction of the biholomorphism seems to rely on some delicate facts about complex varieties. We begin with a well-known automatic continuity result. Recall that a commutative Banach algebra is semi-simple if the Gelfand transform is injective. Lemma 5.1. Let V and W be varieties in B d. Every homomorphism from M V to M W is norm continuous. Proof. The algebras that we are considering are easily seen to be semisimple. A general result in the theory of commutative Banach algebras says that every homomorphism into a semi-simple algebra is automatically continuous (see [11, Prop. 4.2]). Lemma 5.2. Let V and W be varieties in B d and B d, respectively, with d <. Let ϕ : M V M W be an algebra isomorphism. Suppose that λ is an isolated point in W. Then ϕ (ρ λ ) is an evaluation functional at a point in V. Proof. The character ρ λ is an isolated point in M(M W ). (Here is where we need d < ). Since ϕ is a homeomorphism, ϕ (ρ λ ) must also be an isolated point in M(M V ). By Shilov s idempotent theorem (see [9, Theorem 21.5]), the characteristic function χ ϕ (ρ λ ) of ϕ (ρ λ )

14 K.R. DAVIDSON, C. RAMSEY, AND O.M. SHALIT belongs to M V. Now suppose that ϕ (ρ λ ) is in the corona M(M V )\V. Then χ ϕ (ρ λ ) vanishes on V. Therefore, as an element of a multiplier algebra, this means that χ ϕ (ρ λ ) = 0. Therefore χ ϕ (ρ λ ) must vanish on the entire maximal ideal space, which is a contradiction. Thus ϕ (ρ λ ) lies in V. Next we want to show that any algebra isomorphism ϕ between M V and M W must induce a biholomorphism between W and V. This identification will be the restriction of ϕ to the characters of evaluation at points of W. In order to achieve this, we need to make some additional assumption. Our difficulty is basically that we do not have enough information about varieties. In the classical case, if one takes a regular point λ V, takes the connected component of λ in the set of all regular points of V, and closes it up (in B d ), then one obtains a subvariety. Moreover the closure of the complement of this component is also a variety [28, ch.3, Theorem 1G]. However our varieties are the intersections of zero sets of a family of multipliers. Let us say that a variety V is irreducible if for any regular point λ V, the intersection of zero sets of all multipliers vanishing on a small neighbourhood V b ε (λ) is exactly V. We do not know, for example, whether an irreducible variety in our sense is connected. Nor do we know that if we take an irreducible subvariety of a variety, then there is a complementary subvariety as in the classical case. A variety V is said to be discrete if it has no accumulation points in B d. We will resolve this in two situations. The first is the case of a finite union of irreducible varieties and a discrete variety. The second is the case of an isometric isomorphism. In the latter case, the isomorphism will turn out to be completely isometric. This yields a different approach to the results of the previous section. We need some information about the maximal ideal space M(M V ). Recall that there is a canonical projection π into B d obtained by evaluation at [Z 1,..., Z d ]. For any point µ in the unit sphere, π 1 (µ) is the fiber of M(M V ) over µ. We saw in Proposition 3.2 that for λ B d, π 1 (λ) is the singleton {ρ λ }, the point evaluation at λ. The following lemma is analogous to results about Gleason parts for function algebras. However part (2) shows that this is different from Gleason parts, as disjoint subvarieties of V will be at a distance of less than 2 apart. This is because M V is a (complete) quotient of M d, and thus the difference ρ λ ρ µ is the same whether evaluated as functionals on M V or M d. In the latter algebra, λ and ν do lie in the same Gleason part.

OPERATOR ALGEBRAS FOR ANALYTIC VARIETIES 15 Lemma 5.3. Let V be a variety in B d. (1) Let ϕ π 1 (µ) for µ B d. Suppose that ψ M(M V ) satisfies ψ ϕ < 2. Then ψ also belongs to π 1 (µ). (2) If λ and µ belong to V, then ρ µ ρ λ 2r < 2, where r is the pseudohyperbolic distance between µ and λ. Proof. If ψ π 1 (ν) for ν µ in the sphere, then there is an automorphism of B d that takes µ to (1, 0,..., 0) and ν to ( 1, 0,..., 0). Proposition 4.1 shows that composition by this automorphism is a completely isometric automorphism. So we may suppose that µ = (1, 0,..., 0) and ν = ( 1, 0,..., 0). But then ψ ϕ (ψ ϕ)(z 1 ) = 2. Similarly, if ψ = ρ λ for some λ V, then for any 0 < ε < 1, there is an automorphism of B d that takes µ to (1, 0,..., 0) and ν to ( 1 + ε, 0,..., 0). The same conclusion is reached by letting ε decrease to 0. If λ and µ belong to V, then there is an automorphism γ of B d sending λ to 0 and µ to some v := (r, 0,..., 0) where 0 < r < 1 is the pseudohyperbolic distance between λ and µ. Given any multiplier f M V with f = 1, Proposition 2.6 provides a multiplier f in M d so that f V = f and f = 1. In particular, f γ 1 is holomorphic on B d and f γ 1 1. Hence the Schwarz Lemma [27, Theorem 8.1.4] shows that f(µ) f(λ) 1 f(µ)f(λ) = f γ 1 (v) f γ 1 (0) 1 f γ 1 (v) f γ 1 (0) r. Hence ρ µ ρ λ = sup (ρ µ ρ λ )(f) r sup 1 f(µ)f(λ) 2r. f 1 f 1 This provides some immediate information about norm continuous maps between these maximal ideal spaces. Corollary 5.4. Suppose that ϕ is a continuous algebra homomorphism of M V into M W. (1) Then ϕ maps each irreducible subvariety of W into V or into a single fiber of the corona. (2) If ϕ is an isomorphism, and V and W are the disjoint union of finitely many irreducible subvarieties, then ϕ must map W onto V. (3) If ϕ is an isometric isomorphism, then ϕ maps W onto V and preserves the pseudohyperbolic distance.

16 K.R. DAVIDSON, C. RAMSEY, AND O.M. SHALIT Proof. (1) Let W 1 be an irreducible subvariety of W, and let λ be any regular point of W 1. We do not assert that W 1 is connected. Suppose that ϕ (ρ λ ) is a point evaluation at some point µ in B d. Then by Proposition 3.2, µ belongs to V. Since ϕ is norm continuous, by Lemma 5.3 it must map the connected component of λ into a connected component of V. Similarly, suppose that ϕ (ρ λ ) is mapped into a fiber of the corona. Without loss of generality, we may suppose that it is the fiber over (1, 0,..., 0). Since ϕ is norm continuous, by Lemma 5.3 it must map the connected component of λ into this fiber as well. Suppose that there is some point µ in W 1 mapped into V or into another fiber. So the whole connected component of µ is also mapped into V or another fiber. Then the function h = ϕ(z 1 ) 1 vanishes on the component of λ but does not vanish on the component containing µ. This contradicts the fact that W 1 is irreducible. Thus the whole subvariety must map entirely into a single fiber or entirely into V. (2) Suppose that W is the union of irreducible subvarieties W 1,..., W n. Fix a point λ W 1. For each 2 i n, there is a multiplier h i M d which vanishes on W i but h i (λ) 0. Hence h = h 2 h 3 h k W belongs to M W and vanishes on k i=2w i but not on W 1. Therefore ϕ 1 (h) = f is a non-zero element of M V. Suppose that ϕ (W 1 ) is contained in a fiber over a point in the boundary of the sphere, say (1, 0,..., 0). Since Z 1 1 is non-zero on V, we see that (Z 1 1)f is not the zero function. However, (Z 1 1)f vanishes on ϕ (W 1 ). Therefore ϕ((z 1 1)f) vanishes on W 1 and on k i=2w i. Hence ϕ((z 1 1)f) = 0, contradicting injectivity. We deduce that W 1 is mapped into V. By interchanging the roles of V and W, we deduce that ϕ must map W onto V. (3) In the isometric case, we can make use of Lemma 5.3(2) because then ϕ is also isometric. Therefore all of W is mapped by ϕ either into V or into a single fiber. In the latter case, we may suppose that the fiber is over (1, 0,..., 0). Then ϕ(z 1 1) will vanish on all of W, and hence ϕ(z 1 1) = 0, contradicting injectivity. Thus W is mapped into V. Reversing the role of V and W shows that this map is also onto V. The proof of Lemma 5.3(2) actually yields more information, namely that ρ λ ρ µ is a function of the pseudohyperbolic distance r, ρ λ ρ µ = r sup 1 f(µ)f(λ). f 1 In the proof of that lemma we only used that the left hand side is less than or equal to the right hand side, but it is easy to see that

OPERATOR ALGEBRAS FOR ANALYTIC VARIETIES 17 one obtains equality by choosing a particular f. So the fact that the quantities ρ λ ρ µ and sup f 1 1 f(µ)f(λ) are preserved by an isometric isomorphism implies that the pseudohyperbolic distance r is also preserved. Remarks 5.5. (1) In a previous version of this paper, we claimed incorrectly that if ϕ is a surjective continuous homomorphism of M V onto M W, then ϕ must map W into V. This is false, and we thank Michael Hartz for pointing this out. This follows from Hoffman s theory [21] of analytic disks in the corona of H. There is an analytic map L of the unit disk D into the corona of M(H ), mapping onto a Gleason part, with the property that ϕ(h)(z) = h(l(z)) is a homomorphism of H onto itself [17, ch.x 1]. Therefore the map ϕ maps the disk into the corona via L. (2) The main obstacle preventing us from establishing part (2) of the corollary in greater generality is that we do not know that if λ W, then there is an irreducible subvariety W 1 W containing λ and another subvariety W 2 W so that λ W 2 and W = W 1 W 2. As mentioned in the introduction, for any classical analytic variety this is possible [28, ch.3, Theorem 1G]. But our definition requires these subvarieties to be the intersection of zero sets of multipliers. Moreover our proof makes significant use of these functions. So we cannot just redefine our varieties to have a local definition as in the classical case even if we impose the restriction that all functions are multipliers. A better understanding of varieties in our context is needed. (3) Costea, Sawyer and Wick [10] establish a corona theorem for the algebra M d. That is, the closure of the ball B d in M(M d ) is the entire maximal ideal space. This result may also hold for the quotients M V, but we are not aware of any direct proof deducing this from the result for the whole ball. A corona theorem for M V would resolve the difficulties in case (2). The topology on V = B d M(M V ) coincides with the usual one. In particular, each component has closed complement. The corona theorem would establish that every open subset of any fiber is in the closure of its complement. Thus any homeomorphism ϕ of M(M W ) onto M(M V ) must take W onto V. However it is likely that the corona theorem for M V is much more difficult than our problem. Now we can deal with the case in which our variety is a finite union of nice subvarieties, where nice will mean either irreducible or discrete. Theorem 5.6. Let V and W be varieties in B d, with d <, which are the union of finitely many irreducible varieties and a discrete variety.

18 K.R. DAVIDSON, C. RAMSEY, AND O.M. SHALIT Let ϕ be a unital algebra isomorphism of M V onto M W. Then there exist holomorphic maps F and G from B d into C d with coefficients in M d such that (1) F W = ϕ W and G V = (ϕ 1 ) V (2) G F W = id W and F G V = id V (3) ϕ(f) = f F for f M V, and (4) ϕ 1 (g) = g G for g M W. Proof. First we show that ϕ maps W into V. Write W = D W 1 W n where D is discrete and each W i is an irreducible variety. The points in D are isolated, and thus are mapped into V by Lemma 5.2. A minor modification of Corollary 5.4(2) deals with the irreducible subvarieties. Since D is a variety, there is a multiplier k M d which vanishes on D and is non-zero at a regular point λ W 1. Proceed as in the proof of the lemma, but define f = h 2... h n k. Then the argument is completed in the same manner. Reversing the roles of V and W shows that ϕ maps W onto V. We have observed that ϕ (ρ λ ) lies in the fiber over the point F (λ) = (ϕ(z 1 )(λ),..., ϕ(z d )(λ)). Since we now know that ϕ maps W into V, we see (with a slight abuse of notation) that F = ϕ W. In particular, the coefficients of F are multipliers. Thus by Proposition 2.6, each f i is the restriction to W of a multiplier in M d, which we also denote by f i. In particular, each f i is holomorphic on the entire ball B d. Thus (since d < ), F is a bounded holomorphic function of the ball into C d. It may not carry B d into itself, but we do have F (W ) = V. A similar argument applied to ϕ 1 shows that G(V ) W and G V = (ϕ 1 ) V. Since (ϕ 1 ) = (ϕ ) 1, we obtain that G F W = id W and F G V = id V. The last two statements follow as in Lemma 4.2. Remark 5.7. Note that in the above theorem, the map F can be chosen to be a polynomial if and only if the algebra homomorphism ϕ takes the coordinate functions to (restrictions of) polynomials; and hence takes polynomials to polynomials. Likewise, F can be chosen to have components which are continuous multipliers if and only if ϕ takes the coordinate functions to continuous multipliers; and hence takes all continuous multipliers to continuous multipliers. Remark 5.8. When d =, there is no guarantee that the map F constructed in our proof would actually have values in l 2. However if we assume that ϕ is completely bounded, then we can argue as follows.

OPERATOR ALGEBRAS FOR ANALYTIC VARIETIES 19 The row operator Z = [ Z 1 Z 2 Z 3... ] [ is a contraction. Thus ϕ(z) = ϕ(z1 ) ϕ(z 2 ) ϕ(z 3 )... ] is bounded by ϕ cb. By Proposition 2.6, there are functions f i M d so that f i W = ϕ(z i ) and [ M f1 M f2 M f3... ] ϕ cb. In particular, F = [ f 1 f 2 f 3... ] is bounded by ϕ cb in the sup norm. Theorem 5.6 can then be modified to apply in the case d =. However these hypotheses are very strong. Corollary 5.9. Every algebraic automorphism of M d for d finite is completely isometric, and is unitarily implemented. Proof. The previous theorem shows that every automorphism is implemented as composition by a biholomorphic map of the ball onto itself, i.e. a conformal automorphism of B d. Proposition 4.1 shows that these automorphisms are completely isometric and unitarily implemented. Now we consider the isometric case. Theorem 5.10. Let V and W be varieties in B d, with d <. Every isometric isomorphism of M V onto M W is completely isometric, and thus is unitarily implemented. Proof. Let ϕ be an isometric isomorphism of M V onto M W. By Corollary 5.4(3), ϕ maps W onto V and preserves the pseudohyperbolic distance. Let F be the function constructed as in Theorem 5.6. As in Lemma 4.2 and Theorem 5.6, F is a biholomorphism of W onto V and ϕ(h) = h F. After modifying both V and W by a conformal automorphism of the ball, we may assume that 0 belongs to both V and W, and that F (0) = 0. Set w 0 = 0 and choose a basis w 1,..., w k for span W. Let v p = F (w p ) for 1 p k. Suppose that w p = r p. This is the pseudohyperbolic distance to w 0 = 0 = v 0, so v p = r p as well. Write v p /r p = d j=1 c je j. Let h p (z) = z, v p /r p = d j=1 c jz j (z). This is a linear function on V, and thus lies in M V. Since Z is a row contraction, f has norm at most one. Therefore k p := ϕ(h p ) = h p F has norm at most one in M W. Now let w k+1 = w be an arbitrary point in W, and set v k+1 = v = F (w) V. By a standard necessary condition for interpolation [2, Theorem 5.2], the fact that k p 1 means that in particular interpolating at the points w 0,..., w k, w k+1, we obtain ] 0. [ 1 hp(v i )h p(v j ) 1 w i,w j 0 i,j k+1

20 K.R. DAVIDSON, C. RAMSEY, AND O.M. SHALIT In particular, look at the 3 3 minor using rows 0, p, k + 1 to obtain 1 1 1 1 v,v 0 1 1 p 1 w p,w 1 1 v,v p 1 w,w p 1 v,v p/r p ) 2 1 w 2 By the Cholesky algorithm, we find that 1 v,vp 1 w,w p In particular, we obtain v, v p = w, w p for 1 p k. v i, v j = w i, w j for 1 i, j k. = 1. Therefore Therefore there is a unitary operator U acting on C d such that Uw i = v i for 1 i k. Now since w W lies in span{w 1,..., w k }, it is uniquely determined by the inner products w, w i for 1 i k. Since v has the same inner products with v 1,..., v k, we find that Uw = P N v where N = span{v 1,..., v k }. However we also have v = w = Uw = P N v ; whence v = Uw. Therefore F agrees with the unitary U, and hence ϕ is implemented by an automorphism of the ball. So by Proposition 4.1, ϕ is completely isometric and is unitarily implemented. With these results in hand, we may repeat the arguments in [16, Section 11.3] word for word to obtain the following automatic continuity result. Recall that the weak-operator and the weak- topologies on M V coincide by Lemma 3.1. Theorem 5.11. Let ϕ : M V M W, for d <, be a unital algebra isomorphism given by composition: ϕ(h) = h F where F is a holomorphic map of W onto V whose coefficients are multipliers. Then ϕ is continuous with respect to the weak-operator and the weak- topologies. 6. Examples In this section, we examine a possible converse to Theorem 5.6 in the context of a number of examples. What we find is that the desired converse is not always true. That is, suppose that V and W are varieties in B d and F and G are holomorphic functions on the ball satisfying the conclusions of Theorem 5.6. We are interested in when this implies that the algebras M V and M W are isomorphic.

OPERATOR ALGEBRAS FOR ANALYTIC VARIETIES 21 Finitely many points in the ball. Let V = {v 1,..., v n } B d. Then A V = M V and they are both isomorphic to l n = C(V ). The characters are evaluations at points of V. If W is another n point set in B d, then M W is isomorphic to M V. Also, there are (polynomial) maps f : B d C d and g : B d C d which are inverses of one another when restricted to V and W. And if W is an m point set, m n, then obviously M V is not isomorphic to M W, and there also exists no biholomorphism. In this simple case we see that M V = MW if and only if there exists a biholomorphism, and this happens if and only if W = V. Nevertheless, the situation for finite sets is not ideal. Let V and W be finite subsets of the ball, and let F : W V be a biholomorphism. It is natural to hope that the norm of the induced isomorphism can be bounded in terms of the multiplier norm of F. The following example shows that this is not possible. Example 6.1. Fix n N and r (0, 1). Put ξ = exp( 2πi ) and let n and V = {0} {rξ j } n j=1, W = {0} { r 2 ξj } n j=1. The map F (z) = 2z is a biholomorphism of W onto V that extends to an H function of multiplier norm 2. We will show that the norm of the induced isomorphism M V M W, given by f f F, is at least 2 n. Consider the following function in M V : f(0) = 0 and f(rξ j ) = r n for 1 j n. We claim that the multiplier norm of f is 1. By Proposition 2.6, f is the minimal norm of an H function that interpolates f. The function g(z) = z n certainly interpolates and has norm 1. We will show that it is of minimal norm. The Pick matrix associated to the problem of interpolating f on V by an H function of norm 1 is 1 1 1 1 1 r 1 2n 1 r 2n 1 r 2n 1 r 2 ξξ 1 r 2 ξξ 2 1 r 2 ξξ n 1 r 1 2n 1 r 2n 1 r 2n 1 r 2 ξ 2 ξ 1 r 2 ξ 2 ξ 2 1 r 2 ξ 2 ξ n........ 1 1 r 2n 1 r 2 ξ n ξ 1 r 2n 1 r 2n 1 r 2 ξ n ξ 2 1 r 2 ξ n ξ n

22 K.R. DAVIDSON, C. RAMSEY, AND O.M. SHALIT To show that g is the (unique) function of minimal norm that interpolates f, it suffices to show that this matrix is singular. (We are using well known facts about Pick interpolation. See Chapter 6 in [2]). We will show that the lower right principal sub-matrix [ ] n 1 r 2n A = 1 r 2 ξ i ξ j i,j=1 has the vector (1,..., 1) t as an eigenvector with eigenvalue n. It follows that (n, 1, 1,..., 1) t is in the kernel of the Pick matrix. (The matrix A is invertible, so the Pick matrix has rank n). Indeed, for any i, n j=1 1 r 2n 1 r 2 ξ i ξ j = (1 r2n ) = (1 r 2n ) = (1 r 2n ) n j=1 k=0 k=0 j=1 (r 2 ξ i ξ j ) k n r 2k ξ ik ξ jk nr 2mn ξ imn = n 1 r2n = n. 1 r2n We used the familiar fact that n j=1 ξjk is equal to n for k 0 (mod n) and equal to 0 otherwise. Therefore f = 1. Now we will show that f F M W has norm 2 n, where F (z) = 2z. The function f F is given by m=0 f F (0) = 0 and f F ( r 2 ξj ) = r n for 1 j n. The unique H function of minimal norm that interpolates f F is h(z) = 2 n z n. This follows from precisely the same reasoning as above. Therefore the isomorphism has norm at least 2 n. Blaschke sequences. We will now provide an example of two discrete varieties which are biholomorphic but yield non-isomorphic algebras. Example 6.2. Let v n = 1 1/n 2 and w n = 1 e n2 for n 1. Set V = {v n } n=1 and W = {w n } n=1. Both V and W satisfy the Blaschke condition so they are analytic varieties in D. Let B(z) be the Blaschke product with simple zeros at points in W. Define h(z) = 1 e 1 z 1,

OPERATOR ALGEBRAS FOR ANALYTIC VARIETIES 23 and log(1 z) + 1 ( g(z) = log(1 z) Then g, h H and they satisfy 1 B(z) B(0) ). h g W = id W and g h V = id V. However, by the corollary in [22, p.204], W is an interpolating sequence and V is not. Thus the algebras M V and M W cannot be similar by a map sending normalized kernel functions to normalized kernel functions. The reason is that the normalized kernel functions corresponding to an interpolating sequence form a Riesz system, while those corresponding to a non-interpolating sequence do not. In fact, M V and M W cannot be isomorphic via any isomorphism, as we see below. Theorem 6.3. Let V = {v n } n=1 B d, with d <, be a sequence satisfying the Blaschke condition (1 v n ) <. Then M V is isomorphic to l if and only if V is interpolating. Proof. By definition, V is interpolating if and only if M V is isomorphic to l via the restriction map. It remains to prove that if V is not an interpolating sequence, then M V cannot be isomorphic to l via any other isomorphism. Let V be a non-interpolating sequence, and let W be any interpolating sequence. If M V is isomorphic to l, then it is isomorphic to M W. But by Lemma 5.2, this isomorphism must be implemented by composition with a holomorphic map, showing that M V is isomorphic to l via the restriction map. This is a contradiction. Remark 6.4. We require the Blaschke condition to insure that V is a variety of the type we consider, i.e., a zero set of an ideal of multipliers (see [5, Theorem 1.11]). Any discrete variety in D satisfies this condition. Curves. Let V be a variety in B d. If M V is isomorphic to H (D), then by Theorem 5.6 we know that V must be biholomorphic to the disc. To study the converse implication, we shall start with a disc biholomorphically embedded in a ball and try to establish a relationship between the associated algebras M V and its reproducing kernel Hilbert space F V and H (D) and H 2 (D). Suppose that h is a holomorphic map from the disc D into B d such that h(d) = V, and that there exists a holomorphic map g : B d C such that g V = h 1. The following result shows that in many cases, the desired isomorphism exists [3]. See [4, 2.3.6] for a strengthening to planar domains, and a technical correction.

24 K.R. DAVIDSON, C. RAMSEY, AND O.M. SHALIT Theorem 6.5 (Alpay-Putinar-Vinnikov). Suppose that h is an injective holomorphic function of D onto V B d such that (1) h extends to a C 1 function on D, (2) h(z) = 1 if and only if z = 1, (3) h(z), h (z) 0 when z = 1. Then M V is isomorphic to H. Condition (3) should be seen as saying that V meets the boundary of the ball non-tangentially. We do not know whether such a condition is necessary. The authors of [3] were concerned with extending multipliers on V to multipliers on the ball. This extension follows from Proposition 2.6. By the results of Section 4, there is no loss of generality in assuming that h(0) = 0, and we do so. Define a kernel k on D by k(z, w) = k(h(z), h(w)) = 1 1 h(z), h(w). Let H be the RKHS determined by k. Write k w for the function k(, w). The following routine lemma shows that we can consider this new kernel on the disc instead of F V. Lemma 6.6. The map k z k h(z) extends to a unitary map U of H onto F V. Hence, the multiplier algebra Mult(H) is unitarily equivalent to M V. This equivalence is implemented by composition with h: U M f U = M f h for f M V. Proof. A simple computation shows that c i kzi 2 c i c j = 1 h(z i i,j i ), h(z j ) = i c i k h(zi 2 ). So we get a unitary U : H F V. As in the proof of Proposition 4.1, for all f M V we have U M f U = M f h. Our goal in this section is to study conditions on h which yield a natural isomorphism of the RKHSs H and H 2 (D). The first result is that the Szego kernel k z dominates the kernel k z. Lemma 6.7. Suppose that h is a holomorphic map of D into B d. Then for any finite subset {z 1,..., z n } D, [ ] [ ] 1 1. 1 h(z j ), h(z i ) 1 z j z i